Leaked audited financials show a $20.9 billion operating loss on $13 billion in revenue. The loss is shrinking as a share of revenue, and 2030 is the year the company is asking investors to measure it against.
OpenAI generated $13.07 billion in revenue in 2025 and posted a $20.92 billion operating loss. The wire will lead with one of those numbers. The more useful question is why both are true at the same time, and the answer sits in a stack of audited financial statements that the company did not release.
The documents, first obtained by independent journalist Ed Zitron and independently reviewed by the Financial Times, show OpenAI scaling faster than almost any software company in history, and spending even faster. Revenue nearly quadrupled year over year, from $3.7 billion in 2024 to $13.07 billion in 2025. By the end of last year, monthly revenue had grown to nearly $2 billion, per the FT's review of the same set of documents. The operating loss grew with it, from $8.78 billion in 2024 to $20.92 billion in 2025. That is the surface story.
The structural story is the more useful one. OpenAI scaled revenue 3.5 times while scaling total costs 2.4 times. The operating loss, expressed as a share of revenue, fell from 237% in 2024 to 160% in 2025. The absolute number is still enormous, and the company itself is telling investors it does not expect to be profitable until 2030. But the trajectory in the audited documents points somewhere other than either the doom narrative or the bubble narrative.
To see why, follow the money.
R&D is the largest of the three big cost lines, and the one most readers will not recognize as a single payment. It grew from $7.81 billion in 2024 to $19.18 billion in 2025. Of the 2025 figure, $10.59 billion flowed to Microsoft, the company's primary compute and cloud partner, for training the next generation of models. This is the cost the rest of the cost stack is built on top of.
Inference-time compute, the cost of actually running the models for users and API customers, is recorded under cost of revenue. That line grew from $2.65 billion in 2024 to $7.5 billion in 2025. It scales with usage, not with the training cycle, and it is the line that determines whether serving a ChatGPT query is profitable on the margin.
Sales and marketing is the sleeper. It grew from $1.11 billion in 2024 to $5.73 billion in 2025, a more than fivefold jump, as OpenAI built out enterprise and consumer sales capacity to convert its technical lead into recurring revenue.
Put those three lines together and you have most of the gap between revenue and operating loss. R&D plus cost of revenue plus sales and marketing totaled $32.4 billion in 2025 against $13.07 billion in revenue. The remaining operating costs push the operating loss to $20.92 billion.
The honest reading of these documents is that OpenAI is doing two things at once. It is losing more money in absolute terms than at any point in its history, and it is making more progress on the ratio of revenue to total cost than at any point in its history. Both can be true, and both are in the audited statements. The wire will pick one. The numbers, read together, support a more careful picture.
Two caveats matter. These are leaked documents, not a public filing, even though the Financial Times independently reviewed the same set. The specific dollar amounts should be re-verified against the full Ars Technica piece, Zitron's write-up, and FT coverage before being treated as final. The company is also now filing SEC paperwork ahead of an expected initial public offering, which is part of why these figures are surfacing now, and why the 2030 profitability target, OpenAI's own claim to investors, is the benchmark the audited numbers will eventually be measured against.
The next test is the 2026 numbers. If revenue keeps compounding toward the $2 billion monthly run rate and the loss as a share of revenue keeps falling at the pace these documents show, the 2030 claim becomes a model rather than a slogan. If it does not, these documents become the high-water mark.